Planetary rovers have proven to be invaluable assets within space exploration missions, providing profound insights and expanding scientific knowledge. As the frontiers of space exploration continue to expand, there is a growing demand for rovers with advanced capabilities and a higher degree of autonomy. A key aspect of future rover generations involves their ability to autonomously manipulate objects in extraterrestrial environments, particularly in missions involving sample collection, analysis, or return. However, this task presents significant challenges due to the unknown nature of objects and the complex terrains encountered. To address these challenges, this paper investigates the potential of deep learning techniques to enhance autonomous robotic grasping in extraterrestrial environments. The proposed approach introduces an end-to-end grasp estimation system, enabling rovers equipped with robotic arms to autonomously execute grasping actions solely based on visual information provided by its on-board sensors. Transfer learning is employed, harnessing the power of pre-trained deep learning models from computer vision applications and pre-existing public grasping datasets. These models are then finetuned using a self-generated dataset containing objects relevant to manipulation tasks in space exploration missions. To bridge the gap between space exploration and deep learning research, a pipeline is introduced to automatically generate a large, labelled dataset of objects suitable for autonomous grasping in planetary exploration missions. Additionally, a 3D planetary robot simulation environment is developed as the core platform for generating and automatically labelling synthetic custom data, emulating conditions encountered in extraterrestrial environments. Preliminary results demonstrate the promising ability of the system to successfully grasp novel objects based only on RGB-D visual information, achieving a success rate of 85%. This approach enables future rovers equipped with human-like grasping abilities to operate without prior knowledge of target objects, while maintaining resource efficiency, thereby enhancing autonomous on-board decision-making in space missions.


    Zugriff

    Download

    Verfügbarkeit in meiner Bibliothek prüfen


    Exportieren, teilen und zitieren



    Titel :

    Autonomous Robotic Grasping using Deep Learning Methods for Resource Efficient Space Robotics Applications


    Beteiligte:
    Badri, M. (Autor:in) / Gewehr, M. (Autor:in) / Klinkner, S. (Autor:in)

    Erscheinungsdatum :

    2024


    Format / Umfang :

    24 pages



    Medientyp :

    Sonstige


    Format :

    Elektronische Ressource


    Sprache :

    Englisch




    Robot/Balloon Facility for Autonomous Robotic Grasping of Objects in Space

    Sharf, I. / Crymble, D. / International Federation of Automatic Control | British Library Conference Proceedings | 2004


    Review of machine learning in robotic grasping control in space application

    Jahanshahi, Hadi / Zhu, Zheng H. | Elsevier | 2024

    Freier Zugriff

    Grasping learning, optimization, and knowledge transfer in the robotics field

    Pozzi L. / Gandolla M. / Pura F. et al. | BASE | 2022

    Freier Zugriff


    Autonomous Grasping Using Novel Distance Estimator

    Skaldebø, Martin Breivik / Haugaløkken, Bent Oddvar Arnesen / Schjølberg, Ingrid | BASE | 2023

    Freier Zugriff